We propose to develop a novel electrode array (`NeuroGrid') for large-scale recording of spikes and improved miniaturized, multiplexed devices for recording of neural activity in freely behaving rodents, while minimizing the loss of cellular/sub-cellular and temporal resolution. We will develop neuron-size density (10x10 m with 30 m pitch), ultra-conformable (4-m thick) and scalable parylene-based probes (MRI compatible) that can be placed on the brain's surface to record spiking activity of individual neurons and their aggregate activity (local field potentials, LFP). Owing to the scalable nature of the NeuroGrid, we can record from hundreds, potentially several thousands, of neurons from the superficial cortical layers chronically in experimental animals and human patients, i.e., collect orders of magnitude larger samples and for longer time than is possible with current technologies. NeuroGrids will be used for both discovery science to understand neuronal computation that underlies cognition and extracting physiological markers in epileptic patterns that may improve our ability to understand seizure pathogenesis and predict seizure occurrence. The ability of the NeuroGrid to record stable signals for extended time may also support prosthetics. We will test various configurations of the NeuroGrid from 64 to 256 to >1000 recording sites. Multiple approaches of signal multiplexing will be tested to achieve the most compact headstage configuration. In the experimental project, we will perform simultaneous recordings from the depth of the cortex using high-density silicon probes and cell specific activation of layer 1 and 2/3 neurons with optogenetic methods to identify the origin of surface-recorded spikes. We will test and improve various existing methods for unit clustering and will explore novel compressed sensing methods to extract spike data from the NeuroGrid signals. In the clinical part of the project, we will record from both epileptic and 'intact' locations during surgery in patients undergoing diagnostic brain monitoring and examine whether spiking information provides a more reliable marker than LFP. NeuroGrids and know-how will be made available to collaborators and commercialized.